In July, Microsoft plans to open to the public a preview of its machine-learning service, hosted on Windows Azure.
Called Azure ML — and previously known as CloudML — the new service is designed to allow partners and customers to build predictive analytics services using Microsoft-developed tools and templates.
Azure ML got its start as a Microsoft Research project that was codenamed "Passau." Microsoft's developers used the algorithms the company developed for services like Xbox Live and Bing to build out the service. Microsoft Research contributed Passau to Microsoft's Data Platform team. The resulting Azure ML team is headed by Corporate Vice President Joseph Sirosh who joined Microsoft at the end of 2013 from Amazon, where he was Vice President of Amazon's global inventory platform.
Azure ML includes a design studio tool aimed at business analysts; an application programming interface (API) service for deployment; and a software development kit (SDK) for building applications on top of Azure ML. A screen shot of the design studio home screen is posted above. Here's another Microsoft-supplied screen shot of the new tool:
About 100 customers and partners have been using a private version of the Azure ML service in the past few months, according to company officials.
Microsoft's own Microsoft Stores have been using the private preview of Azure ML to lessen the number of fraudulent transactions at their locations. The model built on top of Azure ML scores each transaction as being fraudulen or not. By using the service, the stores have seen a 15 percent to 20 percent reduction in frauds, officials said.
"Users can share models and work spaces with their colleagues," said General Manager Eron Kelly. "There are out-of-the-box algorithms, plus the option of bringing your own. And users can train and optimize the model" before publishing their services.
Machine learning is an increasingly important focus at Microsoft. Microsoft officials loosely define machine learning as "a way of applying historical data to a problem by creating a model and using it to successfully predict future behavior or trends."
Currently, machine learning is often handled on-premises and requires specially trained data scientists and programming languages written specifically for statistical computing. Building machine learning models can take months or longer.
Microsoft officials are not sharing a timeframe as to when they expect Azure ML to come out of preview. They also are not commenting on when they will field Project Sage, Microsoft's recommendation service that's built on top of CloudML/Azure ML.